Efficient epileptic seizure detection by a combined IMF-VoE feature

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Automatic seizure detection from electroencephalogram (EEG) is crucial for on-demand closed-loop therapeutic systems.
• Existing methods may lack the accuracy or real-time capability required for effective seizure management.

Purpose of the Study:

• To propose and validate a new feature, IMF-VoE, for predicting seizure occurrence from EEG signals.
• To evaluate the performance of the IMF-VoE feature in terms of sensitivity, false detection rate, and time delay.

Main Methods:

• Empirical Mode Decomposition (EMD) was used to extract three intrinsic mode functions (IMFs) from EEG signals.
• The variance of the range between the upper and lower envelopes (VoE) was calculated.
• The IMF-VoE feature combined these components to characterize seizure states.

Main Results:

• The IMF-VoE feature achieved 100% sensitivity with a low false detection rate of 0.16 per hour on 80.4 hours of EEG data from 4 patients with 10 seizures.
• Average time delays ranged from 10.7s to 19.4s at different false detection rates.
• The proposed feature demonstrated competitive results compared to recent studies.

Conclusions:

• The IMF-VoE feature effectively distinguishes seizure states from background EEG activity.
• The compact nature of IMF-VoE allows for computationally efficient, real-time seizure detection systems.
• This approach holds promise for advancing on-demand closed-loop therapeutic systems for epilepsy.